Skip to content

Releases: postmates/cernan

snoopy

17 Jan 21:43
@blt blt
Compare
Choose a tag to compare

This release contains a small bug fix for the wavefront sink. In particular, if you specify a non-default bin_width for that sink the configuration value will not persist after the first wavefront flush.

  • b7f9112 :: Correct bin_width goof in Wavefront sink

snoopy

17 Jan 21:42
@blt blt
Compare
Choose a tag to compare

This is the first release in a good long while that will appear on crates. This has been done by moving away from a github-only fork of lua to 'mond'. We've also further improved the async IO situation in cernan, improving the buffering of payloads in the presence of WouldBlock.

  • 8f2520b :: Use blt/mond, not blt/rust-lua53
  • 0671839 :: Statsd Source Reads All Datagrams until WouldBlock
  • 3470b96 :: Fixes Nonblocking TcpStream Operations

snoopy

10 Jan 21:36
@blt blt
Compare
Choose a tag to compare

This release introduces @pulltab's at-least-once updates, meaning that cernan is now able to shut itself down safely when you send it SIGINT or SIGTERM. A handful of experimental sinks and sources have also been added, documentation to appear in the wiki. This commit integrates mio at a deep level into cernan, reducing the total number of OS threads that cernan requires in the common case.

  • 6313149 :: Introduce graceful shutdown
  • a4da52d :: At Least Once Delivery on Graceful Shutdown
  • 3d4c680 :: Correct statsd mapping crash
  • 9629b09 :: Introduction of avro source
  • 3bcd14b :: Introduce Kinesis Sink, Improve Sink Interface
  • 2dc08c4 :: Introduce Event::Raw for kinesis, else
  • 2ef4c4a :: Adds Sync Publication to Avro Source
  • 45e219e :: File Source now tracks truncated files well

snoopy

12 Dec 17:06
@blt blt
Compare
Choose a tag to compare

This release incorporates improvements to the postmates/quantiles library that speed up the processing of Summaries in Prometheus sink, resolving (or greatly reducing) exposition failures in that sink. An InfluxDB formatting bug has also been repaired, which is fancy.

This release contains the first contributions of @doubleyou and @ibotty! :D

snoopy

06 Dec 17:36
@blt blt
Compare
Choose a tag to compare

This release adjusts the Elasticsearch sink to better handle timeouts and clear its internal buffer. Also, we now keep a long-lived sum/count for Prometheus sink Summaries, even despite the windowing.

  • 859ca8a :: Avoid catastrophic failure when an ES request times out (#356)
  • 071328c :: Perform a linear search in ES buffer (#357)
  • 86f753c :: Introduce Telemetry override for 'count' and 'sample_sum' (#354)

snoopy

04 Dec 22:42
@blt blt
Compare
Choose a tag to compare

This release fixes a problem with the prometheus sink's take on summaries. In particular, we were destroying the wrong side of the window, meaning all new values were (eventually) being dumped. This bug caught thanks to @dparton.

  • 23688f6 :: Correct prometheus window summary truncation (#352)

snoopy

04 Dec 22:41
@blt blt
Compare
Choose a tag to compare

This release bundles some small housekeeping -- updating of deps -- with a fix to the Elasticsearch sink. It turns out that in the case of a bulk update succeeding ES will still signal that a few lines have failed. We used to dump those lines. Now, we store them and re-attempt to submit them up to some configurable limit.

  • 3bc5a25 :: Remove ES loglines when they've signaled as OK (#349)
  • 1d32c5e :: Update dependencies via 'cargo update'

snoopy

04 Dec 22:41
@blt blt
Compare
Choose a tag to compare

This commit integrates optional point shedding for telemetry in the Wavefront sink. This is useful to shed telemetry that has gone past the Wavefront backfill cutoff.

  • c824f1c :: Allow wavefront sink to shed points (#347)

snoopy

29 Nov 02:54
@blt blt
Compare
Choose a tag to compare

This release improves the performance of the Prometheus sink when the aggregation server is either slow to hang-up or there are many servers calling in. Previously an exclusive lock was held over the sink aggregator, a real problem when you're trying to ingest and report frequently.

snoopy

27 Nov 20:46
@blt blt
Compare
Choose a tag to compare

This release inaugurates the work of making cernan an at-once delivery kind of system. This is John K's main show. We've also included the feedback of end-users into the prometheus sink, making it behave more like what folks expect and not breaking the bank in doing so:

  • da543d3 :: Introducing chan-signal & Signal Handling (#333)
  • c038a77 :: Remove any unwrap / excepts in prometheus critical path (#338)
  • 9b274cb :: Dramatically improve prometheus write speed, report CPU
  • 428bf5b :: Allow end-users to configure statsd error bounds (#341)
  • c5aaad4 :: Allow an 'enabled' flag to be set for firehose sink (#342)
  • c084e60 :: Windowing for prometheus sink's Summaries (#344)